Ampt said Nov. 11 it is running partner pilots of a battery converter for 800-volt direct-current (VDC) AI data centers, a design intended to handle fast GPU load changes while keeping electricity demand from the grid steadier.
Ampt says storage could cover brief surges, allowing a site’s grid connection to serve its typical load. The company says this approach could also let operators support more computing capacity per megawatt of grid capacity, though it has reported no commercial deployment results.
Ampt lists its rack-mountable 2U module at 80 kilowatts and 100 amps. It says multiple units can be combined for installations serving a rack, a row or a data hall. The converter is designed to maintain 800 VDC output to GPU racks while a shared power bus draws on batteries, grid supply or on-site generation; the battery takes up or releases power as demand shifts.
Ampt lists efficiency as high as 99.2% and says the converter responds to GPU demand shifts in under 1 millisecond. Ampt says the battery-side bus operates between 1,100 and 1,500 VDC, depending on charge, before power is converted for the 800 VDC racks.
The design addresses swings in computing demand. NVIDIA says rack power demand can move between levels associated with about 30% and full utilization within milliseconds. A research paper by authors affiliated with NVIDIA, Microsoft and OpenAI examines how synchronized AI training can produce power fluctuations across racks, data centers and the grid. The equipment is part of wider development around 800 VDC systems, including a LITEON power rack displayed at an NVIDIA MGX exhibit.
Ampt CEO Aaron Gomolak said the system could make data centers “easier on the grid” for surrounding communities. The company has not identified its pilot partners or disclosed the battery’s energy capacity or the converter’s price.
Ampt is targeting general availability in Q3 2027.
